David Miller

We make over 35,000 decisions every day.

The vast majority are easy. But a select few we agonize over for days on end. Sometimes paralyzed with indecision.

I'm no expert, but lately I've been asking myself one question to unlock clarity.

How would I feel if it was taken away?

And here's three common examples:

Moving to a new apartment? → How would you feel if someone else took it?

Changing jobs? → How would you feel if they rescinded the offer?

Scared of a new fitness class? → How will you feel if you skip it?

If the answer to any of these is "I wouldn't care", you might want to reconsider. But if the answer is "I'd be devastated", then it's time to make the leap!

I started a self-taught data science journey 5 years ago.

For most of that time, I let imposter syndrome stop me from sharing my journey in public. But six months ago, I realized I could start on Twitter. Here's what's happened since:

• New job

• 0 to 8k+ followers

• 10+ new data friends

• 1k tweets, 30 threads, and 40 atomic essays

And here's 3 reasons why Twitter is the best place to start learning in public:

Reason #1: It's easy

Compare yourself to veteran data science creators and you'll be intimidated.

Writing a long-form article on Medium → Hard

Filming a 10 minute tutorial on YouTube → Hard

Tweeting once a day → Easy

Start easy to reduce pressure.

Reason #2: It's iterative

Daily tweeting is not a silver bullet solution.

You need to do more if you want to unlock the network, knowledge, and opportunities of your dreams.

But here's what it will give you:

• New ideas

• Self-confidence

• A consistent habit

• Rapid feedback loop

A daily tweet becomes a weekly thread. A weekly thread becomes a monthly newsletter. And a monthly newsletter becomes a full-blown side-project.

Reason #4: It's active

Interesting paradox: most social platforms are not that social.

How often do you see a good conversation in the comments on YouTube, Medium, or GitHub?

On Twitter, you can engage with like-minded folks every day. If you do it for a year, all the people you used to look up to will be your friends.

11 years ago Marc Andreesen claimed that SaaS is eating the world.

One of the reasons SaaS companies are so attractive is how easy they are to measure.

Understand these 6 metrics and you can benchmark the performance of any SaaS company.

Metric #1: Pipeline conversion

The sales process is long and complex.

Pipeline conversion measures the percentage of sales opportunities converted to bookings. Refine this metric by stage and deal size to identify holes in the sales process.

Metric #2: LTV:CAC ratio

Is your product earning more from customers than it cost to acquire them?

The LTV:CAC ratio measures gross profit contribution against costs to acquire. The best companies achieve better than a 3x ratio.

Metric #3: MRR / ARR

The greatest feature of a SaaS company is the predictable nature of revenue.

Monthly recurring revenue, or MRR, measures the total monthly value of active subscriptions. ARR annualizes that number.

Metric #4: DAU / MAU

These days, user engagement is critical even for enterprise SaaS.

Daily to monthly average users ratio measures how often a typical user logs in.

Metric #5: Customer churn

What percentage of your customers are leaving the product every period?

For enterprise SaaS, 10% per year is strong. For a consumer product (like Netflix) that number might be 10% per month.

Metric #6: Dollar-based net retention

I believe net retention is the most important metric for a SaaS company.

DBNR measures the revenue of a cohort of customers relative to its initial size. Keeping this number above 100% proves you can replace lost revenue with upsells.

It's not easy staying motivated.

Especially when your goal is to become a data scientist.

Instead of flooding you with more information, take a break with these 6 quotes:

Quote #1:

“Statistics are like bikinis. What they reveal is suggestive, but what they conceal is vital.” — Aaron Levenstein

Quote #2:

“It’s easy to lie with statistics. It’s hard to tell the truth without statistics.” — Andrejs Dunkels

Quote #3:

“The great thing about predictions is that you can be wrong.” — Chris Wiggins

Quote #4:

“Big data is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it…” — Dan Ariely

Quote #5:

“All models are wrong, but some are useful.” — George E. P. Box

Quote #6:

“Data Scientist (n.): Person who is better at statistics than any software engineer and better at software engineering than any statistician.” — Josh Wills

The biggest mistake I made in learning data science was consuming more than I created.

• Podcasts

• Textbooks

• Online courses

• Medium articles

• YouTube tutorials

Consuming information has its place. But learning by doing (bonus points if it's in public) is turbo for your growth.

Here's 3 ways to do it if you're learning data science:

#1 Complete a project

A project is a forcing function to create new knowledge from the information you consume.

• Writing code

• Building a dashboard

• Writing a research paper

Synthesize what you learned by building something with the knowledge.

#2 Write content

Convey what you learned by writing about it in a concise and interesting way.

• Blog posts

• Twitter threads

• Medium essays

Write about a project you finished for a natural synergy.

#3. Teach the subject

Teaching is the ultimate signal of mastery on a topic.

• Mentorship

• Event speaker

• YouTube videos

Teaching requires you to react and respond to unpredictable questions and issues.

Are you tired of countless online courses that provide the same surface level knowledge?

You need surgical tutorials to help you dive deeper with Python and machine learning. Luckily, there's several channels on YouTube to walk you through advanced topics.

If I were starting out, I would subscribe to these 3:

Abhishek Thakur

Technical deep dives from the world's first Quadruple Grand Master on Kaggle.

• Data science

• Deep learning

• Applied machine learning

Corey Schafer

Tutorials for analysts, engineers, and scientists to advance their skills.

• SQL

• Python

• Software development

sentdex

End-to-end walkthroughs of advanced applications in Python.

• Finance

• Robotics

• Machine learning

• Web development

• Game development

The biggest obstacle I faced in changing my career from accounting to data science?

Practicing daily consistency.

Until I started learning in public on Twitter. Twitter is my magic pill for turning rigorous study into a fun, shared experience. If you're not convinced, here's a simple 5-step process to skyrocket your learning curve.

Step 1: Read about 1 concept

What's the secret to defining how big "one concept" should be?

Start small.

On Day 1 the concept might be Python data types. On Day 100 the concept will be implementing a neural network from scratch.

Step 2: Write down 1 way to apply that knowledge

The blank page is a death sentence for focus and progress.

Prove your organization and diligence by writing down exactly what you plan to do the next day.

Step 3: Go to bed, wake up, do it

If you're like me, you stay up way too late and waste half your morning.

A simple mindset shift that's worked for me is to turn sleep into a hobby. Don't think of a bedtime routine as something you have to do. Start looking forward to a good night's sleep as something you enjoy.

Wake up early, full of energy, and lock in on your task until it's complete.

Step 4: Tweet about it

You don't understand a topic unless you can teach it to a 3rd grader.

Lucky for you - there's an audience of 300+ million eager to learn in simple, bite-sized increments.

Take advantage by synthesizing your knowledge with tweets and images.

Step 5: Repeat

The point of the game is not to win, it's to never stop playing.

Most of us know that consistency is the secret sauce, and yet we still fail to practice it. Twitter made me more consistent, I hope it works for you too.

If you want to run successful data science projects, you can rely on a framework you learned in grade school.

It's called The Scientific Method.

Here are the steps, and how they apply to data science for business.

#1: Observation / question

State the problem.

• Losing customers?

• Designing a new product?

• Spending too much money?

#2: Research topic

To research a potential data science project, you need to do two things.

• Gather and investigate data

• Interview relevant stakeholders

#3: Hypothesis

Now you're becoming a subject matter expert.

Frame your upcoming work with a best guess at the answer.

#4: Experimentation

There are five pieces to a holistic data science experiment:

• Independent variables (features)

• Feature engineering

• Dependent variables (outcome or target variable)

• Model selection

• Model parameters

#5: Analyze results

The most important part of your experiment?

Model evaluation.

#6: Report conclusions

Wrap your work in a deliverable that your stakeholders can understand and consume.

#7: Iterate

Armed with new information, cycle through the process until you reach a conclusion.

I reviewed over 100 resumes in building a data consulting team from 5 to 30 people.

The hard truth?

Most portfolio projects are boring. And 99% of candidates do one of three things:

1. Show up with 0 projects to discuss

2. Create surface-level tutorial projects

3. Abandon independent projects before they're finished

But that means there's good news for you.

You only need 1 (top-notch) project to stand out from the crowd.

How to build a top-notch project?

Use this 5 question framework and avoid boredom and burnout:

1. What problem am I solving?

2. Why is it important to me?

3. What data do I need?

4. What skills will I learn?

5. What is the final product?

There are so many free courses in Python and data science, it's almost impossible to choose the best.

The secret?

It doesn't matter. The most important decision is the decision to start as soon as possible. The second most important decision is to break through learning plateaus from there.

I started with these two and if I had to start over, I'd choose them again in a heartbeat.

#1: edX - Introduction to Computer Science and Programming Using Python

A gentle introduction that blends computer science fundamentals with Python-specific practice.

• Computer science foundations

• Simple algorithms

• Testing and debugging

• Algorithmic complexity

• Data structures

#2: edX - Introduction to Computational Thinking and Data Science

Keep the momentum by diving right into the second course in the sequence.

• Graphs and graph optimization

• Dynamic programming

• Plotting with the pylab package

• Probability, Distributions

• Monte Carlo simulations

• Curve fitting

I started my data science learning journey over five years ago.

It wasn't always efficient or fun. I didn't have an organized plan, so I made inconsistent progress. Here's a snapshot of what that looked like:

• 2 job changes

• 9 online courses

• A $90k graduate degree

• Countless hours on YouTube

• 3 education platform subscriptions

If I had to go back, there's one change I would make to learn faster and have more fun while doing it.

Learn in public with a lean learning approach

I let imposter syndrome stop me from learning in public until 6 months ago.

Until I discovered this framework for writing about what I was learning:

• Tweet

• Atomic essay (this post you're reading)

• Twitter thread

• Medium article

This framework accomplished 3 things:

#1: Daily consistency

Writing a Medium article every day is hard.

Tweeting once a day is easy.

Start easy.

#2: Extra motivation

Learning in public made my journey a shared experience.

Accountability from peers and fans is the best form of motivation I've found yet.

#3: Iterative feedback loop

Test topics with tweets that nail the intersection of being fun to write about and interesting to your peers and audience.

Don't invest in longer writing and projects until you identify something that works.

Over 90% of data science models never make it to production.

This internet fact is misleading - the nature of R&D is that much of the work yields no results and gets abandoned. But the lesson is valid. Too much data science work loses track of the original problem statement.

So I use the same framework on all my data science projects to ensure we stay on track.

Connect your dependent variable to a value equation

And here's a simple example.

Imagine a model for a subscription-based software product that predicts customer churn. Evaluating the accuracy of your model is too simple. Do this instead:

• Segment the population of customers into cohorts

• Calculate annual revenue $ and churn % in each cohort

• Calculate churn reduction $ savings from actions taken due to model predictions

Netflix uses data science to save more than $1 billion per year in customer retention.

And that’s just one of many ways they leverage big data to drive growth.

Here are 5 more:

1. Personalized recommendations

Created by an ensemble of at least 5 models, your home page is unique to your preferences.

Netflix estimates their recommender system drives 80% of streaming decisions.

2. Thumbnail generation and testing

Netflix's cutting edge software process thousands of still images per movie title.

Real-time A/B testing determines which image is driving the highest click-through rates.

3. Movie production location scouting

Movie production is a complex cost optimization problem.

Advanced data science chooses an optimal location by considering scheduling constraints, budget, and scene requirements.

4. Movie editing

Post-production can take hundreds of hours.

Machine learning uses past history to predict when manual intervention is necessary.

5. Streaming quality control

High quality content is worthless without high quality streaming.

Models run day and night to test and re-allocate bandwidth usage.

I hate learning new math concepts - and I got perfect scores on the Math section of the SAT and GRE.

But strong fundamentals in probability, statistics, and linear algebra will help you immensely in your data science career.

Luckily there's a few channels on YouTube who have mastered the art of making complex math exciting and understandable.

1. 3Blue1Brown

3B1B has one goal: make math education entertaining.

How do they do it?

Unparalleled animations, unique perspective, and digestible content.

2. StatQuest w/ Josh Starmer

Josh's philosophy is simple: break seemingly complex topics into core concepts.

The result?

Step-by-step explanations to guide you through the most advanced topics in math and machine learning.

3. Khan Academy

Khan Academy is the pioneer and undisputed champion of free education content.

And they've taken their curriculums and put them on YouTube.

Last year I was rejected by 10 companies before I landed a new job in data science.

The funny part?

I had no problem making it through take-home case studies and live coding interviews. I was getting rejected during final round behavioral interviews - the portion when you need to tell stories about your experience.

So I made three pivotal changes to my preparation that made all the difference.

1. Rehearse out loud

The first recording I played of myself shocked me.

My stories were rambling, non-specific, and dull. So I started rehearsing every story about my past experience at least once a day.

By the time I was ready to try again, I spoke with clarity and excitement.

2. Showcase a portfolio

Interviews are just as exhausting for the interviewer as they are for you.

No matter how engaging the storyteller, it's hard to focus through 30+ minutes of intellectual conversation. I came prepared to showcase a personal data science project live on-screen. This strategy served two purposes: nice change of pace and proof of work.

Win-win.

3. Prepare better questions

Can you believe I once responded "no" when an interviewer asked if I had any questions?

This is a critical moment in every interview. A chance to show your passion for the role. Take the time to research the company and interviewer. You'll have thoughtful questions to ask that set you apart from most other candidates.

Millions of people have started a #100DaysOfCode challenge, but almost no one finishes.

The challenge is meant to accelerate learning by forcing consistency. There are two main rules:

• Code for at least one hour per day

• Tweet daily about your progress

I love the intention, but here's three steps I would take to make the challenge more effective.

Step 1: Reduce the challenge to 30 days

You only need 30 days to create a new habit, and 100 days is too daunting at the beginning.

Step 2: Do the challenge with a buddy

Accountability is much easier to achieve with a peer than with yourself.

Step 3: One day per week, replace coding with writing

Write a weekly blog, article, or newsletter to synthesize your learning and take a break from coding.

Make these changes if you want to be successful with the challenge (and have more fun while doing it)!

Hard truth: learning is not inherently fun.

I wasted my undergraduate years partying. So I dedicated myself to finding an exciting field to learn at the graduate level. That's how I found data science. But even with this newfound passion I was bored again within months.

Until this year, when I discovered a framework that made learning more exciting and more efficient at the same time.

The Feynman Technique

The Feynman Technique is based on mastery through teaching.

Step 1: Identify a topic

Mine is data science. More specifically: building valuable products with data.

What's yours?

Step 2: Explain it to a beginner

Teach the subject to articulate your thoughts and clarify your learning.

Hint: Twitter is a vessel to teach at scale.

Step 3: Reflect, refine, and simplify

Any darn fool can make something complex; it takes a genius to make something simple. - Albert Einstein

Step 4: Organize, review, and convey

Once you can convey your expertise to a broad audience, you've reached mastery.

You have your own framework on the topic.

(Bonus) Step 5: Monetize your expertise

Want to make learning in public even more exciting?

Sell your expertise (through courses, coaching, and software) for a true test of your mastery.

Pair programming is the single most important habit I've developed to improve my coding.

Pair programming is a practice of spending 30-60 minutes with a partner live-coding on-screen. It is standard practice among software development teams, but data scientists rarely borrow this exercise for their work.

If you haven't tried it, here's 3 reasons why you should:

Reason #1: Establish consistency

It's no secret - the key to mastering SQL and Python is through daily practice.

A weekly (or daily) pair programming schedule forces accountability with your peers and will establish a great habit.

Reason #2: Break through plateaus with feedback

There is more to learn about data science than you can ever hope to remember (much less master).

Live feedback from your peers provides insight about techniques you may never learn otherwise.

Reason #3: Write production-quality code

Code reviews suck. And in research-driven data science teams, most people skip them.

This is a horrible long-term practice. Pair programming solves this by putting extra sets of eyes on your code from the minute you write it.

Do you think about learning in public every day and never take action?

I have been learning in public for 6 months now. And I have made more progress on my learning goals over these 6 months than in the previous 3 years combined.

But learning in public is intimidating (it took me years to commit).

So, here's the 1 tip I would give you if you wanted to reduce the pressure to start:

Start by posting one tweet every day.

And here are 3 reasons why:

Reason #1: Short-form writing is easier than long-form writing

I "learned" about learning in public by reading Medium articles about data science.

Analyzing the work of long-form writers was daunting. Tweeting solves this problem with a 280 character limit.

Reason #2: Daily consistency builds a strong habit

Consistently building complex projects and writing long articles seems impossible before you start.

Tweeting is the perfect method to start with baby steps.

Reason #3: Twitter has an active community of data nerds

The hard truth is that no one will see your work if it only lives in a GitHub repository.

But if you tweet about your project while making progress, you can build relationships and an audience to come take a look.